Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Uncertainty representation with extended evidential networks for modeling safety of the intended functionality (Sotif)

: Adee, Ahmad; Munk, Peter; Gansch, Roman; Liggesmeyer, Peter


Baraldi, P.:
30th European Safety and Reliability Conference and 15th Probabilistic Safety Assessment and Management Conference, ESREL/PSAM 2020. E-Proceedings. Online resource : 01 - 06 November 2020, Venice, Italy
Singapore: Research Publishing, 2020
ISBN: 978-981-14-8593-0
European Safety and Reliability Conference (ESREL) <30, 2020, Online>
Probabilistic Safety Assessment and Management Conference (PSAM) <15, 2020, Online>
European Commission EC
H2020; 812788; SAS
Safer Autonomous Systems
Fraunhofer IESE ()
Autonomous vehicle safety; Bayesian networks; Dempster-Shafer Theory; Dependability; Evidential networks; Safety of the intended functionality; SOTIF

Highly automated driving (HAD) vehicles are complex and safety critical systems. They are deployed in an intricate environment which undergoes continual changes. Complexity of these systems as well as sensing and understanding the operational environment results in uncertainties, which needs to be addressed for the safety of HAD vehicles. Ongoing standardization activities (ISO/PAS 21448) to provide Safety of the Intended Functionality (SOTIF) of HAD vehicles intend to address these issues. As part of the SOTIF argumentation, we propose a novel modeling method to represent uncertainty of the system and the environment as well as the propagation of uncertainty through the system. In our previous work, we classified three types of uncertainty, namely aleatory, epistemic and ontological for this purpose. In this paper, we provide multiple plausibility functions of Dempster-Shafer Theory to fully assimilate the representation of ontological uncertainty along with epistemic and aleatory. We implement our proposed method using a commercial Bayesian Network tool. We show the application of our method with a perception classification use case.